Muscle force calculation method and system driven by multi-source data

Abstract

The invention discloses a muscle force calculation method and system driven by multi-source data in the technical field of biomechanics, and the method comprises the steps: employing a motion capture system to collect first training data when a subject carries out target motion, and synchronously collecting second training data corresponding to the first training data, the first training data is kinematics data of the subject, and the second training data is dynamics data of the subject; and performing noise reduction processing on the first training data and the second training data to obtain first processing data and second processing data. According to the method, the target model is constructed by collecting the kinematics data and the dynamics data of the subject, rapid calculation of the muscle strength in the movement process of the user is achieved through the target model, the calculation speed of the muscle strength can be effectively increased, the dynamics modeling and data iteration steps are reduced, computing power resource consumption of computer equipment can be reduced, and the user experience is improved. The reasoning calculation time of the target data is effectively shortened, and timely feedback of the target data is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of biomechanics technology, and in particular to a multi-source data driven muscle force calculation method and system. Background Art

[0002] Muscle strength refers to the ability of muscles to generate a certain amount of tension to maintain posture, initiate, or control movement under load on the musculoskeletal system. This refers to the explosive force exerted by muscle contraction. Muscle strength encompasses both explosive power and endurance. Explosive power refers to the maximum force generated by muscle contraction within a short period of time, while endurance refers to the ability of muscles to sustain isometric or multiple isotonic contractions of a certain intensity during exercise. Measuring and analyzing muscle strength data is of great value and necessity for users involved in rehabilitation medicine, exercise science, and human-computer interaction.

[0003] Traditional muscle force calculation methods rely primarily on multibody dynamics simulation and optimization algorithms. This process typically involves collecting human kinematic data using motion capture equipment in a laboratory setting. Combined with dynamic data collected simultaneously during movement, a model of the human musculoskeletal system is constructed using multibody dynamics equations. Static or dynamic optimization algorithms are then used to calculate the muscle forces for each muscle. This approach addresses the issue of muscle redundancy in human physiology through optimization and has been validated to be both reasonable and accurate. While data-driven approaches can directly map kinematic data to muscle forces, bypassing the complex simulation process, these methods are primarily for scientific research purposes. Furthermore, data-driven approaches rely on infrared motion capture equipment and high-precision laboratory equipment for muscle electrical signals, resulting in cumbersome procedures, limited space, and high costs. Furthermore, the accuracy of data-driven approaches is critically dependent on data volume. Muscle force calculation requires individual kinematic data collection for simulation, a process that consumes significant human and material resources and limits the data size available for model training.

[0004] Therefore, existing muscle force calculation methods often suffer from low computational efficiency due to the large amount of time and computing resources consumed. In addition, the existing multi-body dynamics equation solution and optimization process require a large number of iterative calculations, and a single analysis can take up to several minutes or even hours. In practical applications, they can only be used for post-processing analysis, and are difficult to meet the force feedback requirements of rehabilitation clinical assessment and muscle movement training. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-source data-driven muscle force calculation method and system, which can realize rapid calculation of muscle force based on kinematic data. It has the characteristics of simple structure, easy use, and strong site adaptability, and is suitable for rehabilitation medicine, sports training or human-computer collaboration scenarios.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a multi-source data-driven muscle force calculation method, comprising using a motion capture system to collect first training data of a subject performing a target movement, and simultaneously collecting second training data corresponding to the first training data, wherein the first training data is the subject's kinematic data, and the second training data is the subject's kinetic data;

[0008] Performing noise reduction processing on the first training data and the second training data respectively to obtain first processed data and second processed data;

[0009] performing inverse kinematics processing on the first processed data based on a musculoskeletal multi-body dynamics simulation algorithm to obtain third processed data, and calculating fourth processed data using an inverse dynamics and static optimization method in combination with the second processed data, wherein the third processed data is joint angle data of the subject, and the fourth processed data is muscle force data of the subject;

[0010] performing matching processing on the third processed data and the fourth processed data to obtain fifth data, wherein the fifth data is a structured joint angle data-muscle force data paired dataset with consistent time range and matched frame rate;

[0011] The fifth data is divided into a training set, a validation set, and a test set, and the fifth data is normalized and time series feature sliced, and a deep neural network architecture is used for training to obtain a target model;

[0012] The target model is deployed on a wearable device to collect first user data, obtain the user's joint angle data, input the user's joint angle data into the target model, and the target model outputs target data corresponding to the user's joint angle data through a model mapping method, wherein the first user data is the user's kinematic data, and the target data is the user's muscle force data.

[0013] As a further solution of the present invention: the collecting of first training data when the subject performs the target movement includes:

[0014] When the motion capture system is an infrared motion capture device, data of three-dimensional marker point trajectories of the subject performing the target motion is collected.

[0015] As a further solution of the present invention: the collecting of first training data when the subject performs the target movement includes:

[0016] When the motion capture system is an inertial positioning motion capture device, inertial motion data of the subject performing the target motion is collected.

[0017] As a further solution of the present invention: the synchronous acquisition of second training data corresponding to the first training data includes:

[0018] A force plate was used to collect the force data exerted by the subjects on the support surface of the force plate.

[0019] As a further solution of the present invention: the dynamic data is reaction force data of the support surface facing the subject.

[0020] As a further solution of the present invention: performing noise reduction processing on the first training data and the second training data respectively to obtain first processed data and second processed data includes:

[0021] Filtering the first training data and the second training data based on a filtering algorithm to obtain first filtered data and second filtered data;

[0022] An interpolation algorithm is used to perform interpolation processing on the first filtered data and the second filtered data respectively, and first processed data and second processed data are obtained correspondingly, wherein the first processed data is smooth kinematic data and the second processed data is smooth dynamic data.

[0023] As a further solution of the present invention: matching the third processed data and the fourth processed data to obtain the fifth data includes:

[0024] Using a data cleaning method, the third processed data and the fourth processed data are cleaned respectively to obtain first cleaned data and second cleaned data;

[0025] Using an interpolation method, interpolating the first cleaned data and the second cleaned data respectively to obtain first interpolated data and second interpolated data;

[0026] The first interpolation data and the second interpolation data were aligned using the timestamp alignment method to obtain a structured joint angle data-muscle force data paired dataset with consistent time range and matching frame rate.

[0027] In the second aspect, a multi-source data-driven muscle force calculation system is provided, which includes a data acquisition module, a data denoising module, a mechanical calculation module, a data preprocessing module, a model building module and a model deployment module. The data acquisition module is configured to use a motion capture system to collect first training data when the subject performs a target movement, and synchronously collect second training data corresponding to the first training data, wherein the first training data is kinematic data and the second training data is dynamic data; the data denoising module is configured to perform denoising processing on the first training data and the second training data respectively to obtain first processed data and second processed data; the mechanical calculation module is configured to perform inverse kinematic processing on the first processed data based on a musculoskeletal multi-body dynamics simulation algorithm to obtain third processed data, and use inverse dynamics and static optimization methods combined with the second processed data to calculate the fourth processed data, wherein the third processed data is used to train the target model The fourth processed data is the muscle force data for training the target model; the data preprocessing module is configured to match the third processed data and the fourth processed data to obtain the fifth data, wherein the fifth data is a structured joint angle data-muscle force data pairing data set with consistent time range and frame rate matching; the model building module is configured to divide the fifth data into a training set, a validation set and a test set, and perform normalization and time series feature slicing on the fifth data, and adopt a deep neural network architecture to train to obtain the target model; the model deployment module is configured to deploy the target model to the wearable device to collect the first user data, obtain the user's joint angle data, input the user's joint angle data into the target model, and the target model outputs the target data corresponding to the user's joint angle data through a model mapping method, wherein the first user data is the user's kinematic data, and the target data is the user's muscle force data.

[0028] In a third aspect, a computer device is provided, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the multi-source data-driven muscle force calculation method as described in the above scheme is performed.

[0029] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the multi-source data-driven muscle force calculation method as described in the above scheme is executed.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. In the present invention, the muscle force label data of the subject during the movement is obtained by collecting the kinematic data and dynamic data of the subject and combining the inverse kinematics and dynamics methods. The target model is obtained by adopting the mapping relationship between the angle data of the joint after timestamp alignment and the muscle force data after noise reduction. This can effectively improve the calculation speed of muscle force, reduce the dynamic modeling and data iteration steps in the traditional method, reduce the computing power resource consumption of computer equipment, effectively shorten the inference calculation time of target data, and facilitate timely feedback of data.

[0032] 2. In the present invention, the user's kinematic data is obtained by using a wearable portable device, and the user's muscle force data is obtained through target model mapping. On the one hand, the user only needs to wear a wearable motion capture device, without the need for additional muscle electrical signal sensors or sticky markers, which can significantly reduce the complexity and usage cost of the data acquisition equipment. At the same time, the model can be deployed on edge devices or portable computing units, can be applied to non-laboratory environments, and has good scene adaptability.

[0033] 3. In the present invention, by using joint angle data as the input for muscle force calculation, multi-source data can be accepted during training, including data collected by infrared motion capture devices based on marker points or inertial positioning motion capture devices. By expanding the magnitude of the training data set, it is beneficial to improve the generalization of model predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A diagram showing the steps of the muscle force calculation method of the present invention;

[0035] Figure 2 This is a module diagram of the muscle force calculation system of the present invention;

[0036] Figure 3 A diagram of a computer device according to the present invention.

[0037] In the figure: 1. Data acquisition module; 2. Data denoising module; 3. Mechanical calculation module; 4. Data preprocessing module; 5. Model building module; 6. Model deployment module. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Example:

[0040] See also Figure 1In an embodiment of the present invention, a multi-source data-driven muscle force calculation method is provided, including a multi-body dynamics simulation calculation method and a method for establishing a target model using a deep neural network. The muscle force calculation method provided by this embodiment can realize rapid calculation of muscle force based on kinematic data. It has the characteristics of simple structure, easy use, and strong site adaptability. It is suitable for multiple scenarios such as rehabilitation medicine, sports training, and human-computer collaboration, and has good practicality.

[0041] Specifically, the following steps are included:

[0042] S1: using a motion capture system to collect first training data of a subject performing a target movement, and simultaneously collecting second training data corresponding to the first training data, wherein the first training data is kinematic data of the subject, and the second training data is kinetic data of the subject;

[0043] S2: performing noise reduction processing on the first training data and the second training data respectively to obtain first processed data and second processed data;

[0044] S3: performing inverse kinematics processing on the first processed data based on a musculoskeletal multi-body dynamics simulation algorithm to obtain third processed data, and combining the second processed data with an inverse dynamics and static optimization method to calculate fourth processed data, wherein the third processed data is joint angle data of the subject, and the fourth processed data is muscle force data of the subject;

[0045] S4: performing matching processing on the third processed data and the fourth processed data to obtain fifth data, wherein the fifth data is a structured joint angle data-muscle force data paired dataset with consistent time range and matching frame rate;

[0046] S5: Divide the fifth data into a training set, a validation set, and a test set, normalize the fifth data and perform time series feature slicing, and use a deep neural network architecture to train and obtain a target model;

[0047] S6: Deploy the target model to the wearable device to collect the first user data, obtain the user's joint angle data, input the user's joint angle data into the target model, and the target model outputs target data corresponding to the user's joint angle data through a model mapping method, wherein the first user data is the user's kinematic data, and the target data is the user's muscle force data.

[0048] Preferably, the subjects are 10 healthy adult volunteers aged 23-30 years, 160-180 cm in height, and 50-75 kg in weight, and the subjects have no sports injuries or neurological diseases.

[0049] Preferably, the target motion includes a first motion, a second motion, a third motion, a fourth motion and a fifth motion;

[0050] The first exercise was a free gait exercise lasting 3 minutes;

[0051] The second exercise is a standing-sitting-standing cycle for 5 consecutive times;

[0052] The third exercise is 10 repetitions of ankle dorsiflexion and plantar flexion;

[0053] The fourth exercise is 5 squats;

[0054] The fifth exercise is to raise the single leg forward 5 times for each leg.

[0055] Preferably, the kinematic data is first training data when the subject performs the target movement, and is reaction force data of the support surface facing the subject.

[0056] Preferably, synchronously collecting the second training data corresponding to the first training data includes:

[0057] A force platform is used to collect the reaction force data of the subject on the support surface, wherein the reaction force data is the second training data. The force platform is a Bertec force platform, and the ground reaction force is synchronously collected by adopting the point-sticking scheme of the Plug-in-Gait extended model.

[0058] Preferably, performing noise reduction processing on the first training data and the second training data respectively to obtain first processed data and second processed data includes:

[0059] Filtering the first training data and the second training data based on a filtering algorithm to obtain first filtered data and second filtered data;

[0060] An interpolation algorithm is used to perform interpolation processing on the first filtered data and the second filtered data respectively, and first processed data and second processed data are obtained correspondingly, wherein the first processed data is smooth kinematic data and the second processed data is smooth dynamic data.

[0061] Preferably, the inverse kinematics analysis is to calculate the angle sequence of each joint, use the error function to minimize the distance between the marker point and the virtual marker point of the model, and extract the joint angle data of the subject's shoulder, elbow, hip, knee, and ankle joints from the output results of the inverse kinematics analysis. The obtained joint angle data is in degrees (°) with a sampling rate of 120Hz, and the muscle force data is in Newtons (N) with a sampling rate of 1000Hz.

[0062] Preferably, the inverse dynamics analysis calculates the joint torque at each frame, with the input being the joint angle and the ground reaction force.

[0063] Preferably, the static optimization analysis is to calculate muscle force, the objective function is to minimize the total muscle strain energy, and the time series data of the rectus femoris, gastrocnemius, tibialis anterior and gluteus maximus are extracted from the output results of the static optimization analysis.

[0064] Preferably, matching the third processed data and the fourth processed data to obtain the fifth data includes:

[0065] Using a data cleaning method, the third processed data and the fourth processed data are cleaned respectively to obtain first cleaned data and second cleaned data;

[0066] Using an interpolation method, interpolating the first cleaned data and the second cleaned data respectively to obtain first interpolated data and second interpolated data;

[0067] The first interpolation data and the second interpolation data were aligned using the timestamp alignment method to obtain a structured joint angle data-muscle force data paired dataset with consistent time range and matching frame rate.

[0068] Normalization: Both joint angles and muscle forces are normalized to their maximum and minimum values:

[0069]

[0070] The floating window length is set to 10 frames, and sequence samples are generated for time-dependent learning.

[0071] The dimension of each input sample is (10,10), corresponding to 10 input features (five pairs of joint angles) and 4 muscle forces.

[0072] Preferably, an LSTM network is used as the architecture of the target model, and the target model structure includes an input layer, a first LSTM layer, a second LSTM layer, a Dropout layer, and a fully connected layer;

[0073] The dimension of the input layer is (10,10), the first LSTM layer includes 128 hidden units, the second LSTM layer has 64 units, the probability coefficient of the Dropout layer is 0.2, and the output dimension of the fully connected layer is 4, corresponding to 4 muscles.

[0074] The input layer receives a preprocessed dataset consisting of joint angles and muscle forces. The first LSTM layer uses 128 hidden units to perform preliminary feature extraction and temporal dependency modeling on the input data. The second LSTM layer further performs in-depth feature extraction on the output of the first LSTM layer, reducing information redundancy while preserving the key temporal characteristics of the data. The Dropout layer uses regularization strategies to effectively prevent overfitting during training and improve the model's generalization capabilities. Finally, the fully connected layer maps the output of the second LSTM layer to four output nodes, corresponding to the predicted muscle force values of the four muscles.

[0075] Preferably, the activation function of the target model is a ReLU function.

[0076] Preferably, the loss function of the target model adopts mean square error:

[0077]

[0078] The training uses the Adam optimizer, a learning rate of 0.001, a batch size of 64, and 100 epochs.

[0079] The following metrics were used to evaluate the model’s predictive ability:

[0080] Root Mean Square Error (RMSE):

[0081]

[0082] Mean Absolute Error (MAE):

[0083]

[0084] Pearson correlation coefficient (R):

[0085]

[0086] like Figure 2 As shown, this embodiment provides a multi-source data-driven muscle force calculation system, which includes a data acquisition module 1, a data denoising module 2, a mechanical calculation module 3, a data preprocessing module 4, a model building module 5, and a model deployment module 6. The data acquisition module 1 is configured to use a motion capture system to collect first training data when a subject performs a target movement, and simultaneously collect second training data corresponding to the first training data. The data denoising module 2 is configured to perform denoising on the first training data and the second training data, respectively, to obtain first processed data and second processed data.

[0087] The system uses the OpenSim 4.5 platform to provide a musculoskeletal multi-body dynamics simulation algorithm. The mechanical calculation module 3 is a quad-core CPU and has a Gait2392 model with a deep neural network architecture. The Gait2392 model contains 23 degrees of freedom and 92 muscles, covering the key muscle groups of the rectus femoris, tibialis anterior, gastrocnemius, and gluteus maximus. The static posture data extracted through the calibration test is used to individually adjust the model length using the Scaling tool. The calibration process includes a "static standing" posture to align the skeleton with the reflective markers. Based on the mechanical calculation module 3, the first processed data is subjected to inverse kinematics processing to obtain the third processed data. The fourth processed data is calculated using inverse dynamics and static optimization methods combined with the second processed data. Among them, the third processed data is the subject's joint angle data, and the fourth processed data is the subject's muscle force data. The simulation analysis based on the multi-body dynamics musculoskeletal model is completed to obtain the subject's muscle force data.

[0088] The data preprocessing module 4 is configured to perform matching processing on the third processed data and the fourth processed data to obtain fifth data, wherein the fifth data is a structured joint angle data-muscle force data pairing dataset with a consistent time range and a matching frame rate;

[0089] The model building module 5 is configured to perform model learning and training based on a recurrent neural network deep learning architecture using the fifth data, and obtains a target model with joint angle as input and muscle force as output through model tuning;

[0090] The model deployment module 6 is used to deploy the target model on the wearable device to collect the first user data. This can be used to collect the first user data in a non-laboratory environment, thereby improving the adaptability of the target model to different occasions.

[0091] Preferably, the muscle force calculation system also includes a communication module, which is used to transmit the calculated muscle force data to an external device or a cloud platform for use in rehabilitation assessment, training feedback, robot control and other systems.

[0092] Preferably, the muscle force calculation system also includes a display module, which is integrated into a mobile terminal or a terminal computer and is used to display the muscle force calculation results in real time in graphical or numerical form, supporting user queries and result exports.

[0093] Preferably, the data acquisition module is one or more infrared motion capture devices, which are used to collect first training data when the subject performs the target motion, and synchronously collect second training data corresponding to the first training data and send it to the data denoising module. The data denoising module performs denoising on the collected data, and the data denoising module sends the denoised data to the mechanical calculation module for inverse kinematics processing, and sends the processed fourth processed data to the data preprocessing module. The data preprocessing module matches the third processed data and the fourth processed data to obtain fifth data. The model building module divides the fifth data into a training set, a validation set and a test set, and performs normalization and time series feature slicing on the fifth data. A deep neural network architecture is used for training to obtain a target model. The model deployment module deploys the target model to the wearable device to collect first user data, obtain the user's joint angle data, input the user's joint angle data into the target model, and the target model outputs target data corresponding to the user's joint angle data through a model mapping method to obtain the user's muscle force data.

[0094] Preferably, the infrared motion capture device is a 200 Hz Vicon optical motion capture system.

[0095] In this embodiment, the method for the infrared motion capture device to collect the first training data when the subject performs the target movement is to collect the data of the three-dimensional marker point trajectory of the subject performing the target movement, and the infrared motion capture device is used to collect the motion data of the target human segment in the corresponding model training. The mechanical calculation module is based on the experimental data and performs inverse kinematics analysis to calculate the angles of each joint, and then performs multi-body dynamics simulation to calculate the value of each muscle force. The joint angle data and the muscle force data are matched and processed to obtain a structured joint angle data-muscle force data pairing data set with a consistent time range and a matching frame rate. By using the infrared motion capture device, the accuracy of the first training data when the subject performs the target movement can be met, ensuring the accuracy of data collection. The joint angle data and the muscle force data of the subject are matched and processed to obtain a structured joint angle data-muscle force data pairing data set with a consistent time range and a matching frame rate, which is conducive to completing the training of the target model.

[0096] In this embodiment, a computing module is used to read the first training data of the subject performing the target motion collected by the infrared motion capture device.

[0097] In this embodiment, the muscle force calculation method also includes a method for deploying the target model in an actual application scenario. The deployment method includes the following steps:

[0098] The target model is deployed on a wearable device through 12 fixed on the user's torso, upper arm, forearm, thigh, calf, and foot to collect data on the user's shoulder, elbow, hip, knee, and ankle performing target motion. The user's joint angle data is obtained through an open source multi-threaded inverse kinematics algorithm, and the user's joint angle data is input into the target model. The target model outputs target data corresponding to the user's joint angle data through a model mapping method, wherein the first user data is the user's kinematic data, and the target data is the user's muscle force data.

[0099] These methods and systems offer the advantages of high sampling rates, low latency, and freedom from light constraints, enabling them to accurately capture subtle changes in a subject's movements. The computational module utilizes advanced multi-threading technology to ensure efficient operation even when processing large amounts of sensor data. Inverse kinematics analysis, using complex mathematical algorithms to convert sensor-captured motion data into joint angles, is key to linking motion data with muscle force prediction.

[0100] In this embodiment, the pre-deployed target model is constructed based on a deep learning or machine learning algorithm, which can learn the mapping relationship between joint angles and muscle forces from a large amount of training data. When the calculation module inputs the joint angle data into the target model, the target model processes the input data and outputs the corresponding muscle force value. The obtained muscle force value can reflect the force of each muscle when the subject performs the target movement, and can also provide data support for subsequent rehabilitation medicine and sports training.

[0101] like Figure 3 As shown, this embodiment provides a computer device including a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, a multi-source data-driven muscle force calculation method as described above is performed.

[0102] This embodiment further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the multi-source data-driven muscle force calculation method as described above is executed.

[0103] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A multi-source data driven muscle force calculation method, characterized in that: include: Using a motion capture system to collect first training data of a subject performing a target movement, and simultaneously collecting second training data corresponding to the first training data, wherein the first training data is kinematic data of the subject, and the second training data is kinetic data of the subject; Performing noise reduction processing on the first training data and the second training data respectively to obtain first processed data and second processed data; performing inverse kinematics processing on the first processed data based on a musculoskeletal multi-body dynamics simulation algorithm to obtain third processed data, and calculating fourth processed data using an inverse dynamics and static optimization method in combination with the second processed data, wherein the third processed data is joint angle data of the subject, and the fourth processed data is muscle force data of the subject; performing matching processing on the third processed data and the fourth processed data to obtain fifth data, wherein the fifth data is a structured joint angle data-muscle force data paired dataset with consistent time range and matched frame rate; The fifth data is divided into a training set, a validation set, and a test set, and the fifth data is normalized and time series feature sliced, and a deep neural network architecture is used for training to obtain a target model; The target model is deployed on a wearable device to collect first user data, obtain the user's joint angle data, input the user's joint angle data into the target model, and the target model outputs target data corresponding to the user's joint angle data through a model mapping method, wherein the first user data is the user's kinematic data, and the target data is the user's muscle force data.

2. The multi-source data-driven muscle force calculation method according to claim 1, characterized in that: The collecting of first training data when the subject performs the target movement includes: When the motion capture system is an infrared motion capture device, data of three-dimensional marker point trajectories of the subject performing the target motion is collected.

3. The multi-source data-driven muscle force calculation method according to claim 1, characterized in that: The collecting of first training data when the subject performs the target movement includes: When the motion capture system is an inertial positioning motion capture device, inertial motion data of the subject performing the target motion is collected.

4. The multi-source data-driven muscle force calculation method according to claim 2 or 3, characterized in that: The synchronously collecting the second training data corresponding to the first training data includes: A force plate was used to collect the force data exerted by the subjects on the support surface of the force plate.

5. The multi-source data-driven muscle force calculation method according to claim 4, characterized in that: The dynamic data is reaction force data of the support surface acting on the subject.

6. The multi-source data-driven muscle force calculation method according to claim 5, characterized in that: The performing noise reduction processing on the first training data and the second training data respectively to obtain first processed data and second processed data includes: Filtering the first training data and the second training data based on a filtering algorithm to obtain first filtered data and second filtered data; An interpolation algorithm is used to perform interpolation processing on the first filtered data and the second filtered data respectively, and first processed data and second processed data are obtained correspondingly, wherein the first processed data is smooth kinematic data and the second processed data is smooth dynamic data.

7. The multi-source data-driven muscle force calculation method according to claim 6, characterized in that: The matching processing of the third processed data and the fourth processed data to obtain the fifth data includes: Using a data cleaning method, the third processed data and the fourth processed data are cleaned respectively to obtain first cleaned data and second cleaned data; Using an interpolation method, interpolating the first cleaned data and the second cleaned data respectively to obtain first interpolated data and second interpolated data; The first interpolation data and the second interpolation data were aligned using the timestamp alignment method to obtain a structured joint angle data-muscle force data paired dataset with consistent time range and matching frame rate.

8. A multi-source data driven muscle force calculation system, characterized in that: The system comprises: a data acquisition module, the data acquisition module being configured to use a motion capture system to acquire first training data of a subject performing a target movement, and to simultaneously acquire second training data corresponding to the first training data, wherein the first training data is kinematic data and the second training data is kinetic data; a data denoising module, the data denoising module being configured to perform denoising processing on the first training data and the second training data respectively to obtain first processed data and second processed data; a mechanical calculation module, the mechanical calculation module being configured to perform inverse kinematics processing on the first processed data based on a musculoskeletal multi-body dynamics simulation algorithm to obtain third processed data, and to calculate fourth processed data using an inverse dynamics and static optimization method in combination with the second processed data, wherein the third processed data is joint angle data for training a target model, and the fourth processed data is muscle force data for training the target model; a data preprocessing module, the data preprocessing module being configured to perform matching processing on the third processed data and the fourth processed data to obtain fifth data, wherein the fifth data is a structured joint angle data-muscle force data paired dataset having a consistent time range and a matching frame rate; a model building module, the model building module being configured to divide the fifth data into a training set, a validation set, and a test set, perform normalization processing and time series feature slicing on the fifth data, and train the target model using a deep neural network architecture; A model deployment module is configured to deploy the target model onto a wearable device to collect first user data, obtain the user's joint angle data, input the user's joint angle data into the target model, and the target model outputs target data corresponding to the user's joint angle data through a model mapping method, wherein the first user data is the user's kinematic data, and the target data is the user's muscle force data.

9. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the multi-source data-driven muscle force calculation method according to any one of claims 1 to 7 is executed.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the multi-source data-driven muscle force calculation method according to any one of claims 1 to 7 is executed.

Images